| Comparison of the Performance of Different Machine Learning Algorithms in Qualitative Classification of Orange Fruit Based on Image Processing Information |
| Paper ID : 1240-NICAME1402 |
| Authors: |
|
Hemad Zareiforoush *1, Adel Bakhshipour1, Shahriar Ramezanpour13772, Sina Kashaf2 1Department of Biosystems Engineering, Faculty of Agricultural Sciences, University of Guilan, Rasht, Iran 2Department of Biosystems Engineering, Faculty of Agricultural Sciences, University of Guilan, Rasht, Iran. |
| Abstract: |
| In this research, computer vision method and different machine learning algorithms were used to classify different quality grades of orange fruit based on mechanical damage and pest damage indicators. At first, the images of three different quality levels of orange fruit, including healthy, damaged, and infested, were acquired. After image pre-processing, different color, shape, and texture features were extracted from the images of orange fruit samples and used to develop machine learning algorithms. Five important machine learning algorithms including Bayesian Networks (BN), Multilayer Perceptron (MLP) Neural Networks, Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), and Decision Tree (DT) were used for the qualitative classification of orange fruit. In the case of not using the feature reduction algorithm, the DT model of LMT type with the accuracy and RMSE indices equal to 93.33% and 0.2040, respectively, had the best performance at the evaluation stage in identifying the quality grade of orange fruit. By applying the Correlation-based Feature Selection (CFS) algorithm to data, 8 features from 68 primary features were selected as optimal features. In this case, the LMT type DT model had the best performance in the classification of different qualitative degrees of orange fruit based on the image information with accuracy and RMSE values equal to 95.56% and 0.1931, respectively, in the evaluation phase, and after that the MLP model, with two neurons in the hidden layer, had a better performance compared to the other used models with accuracy and RMSE values equal to 92.24% and 0.2205, respectively. |
| Keywords: |
| Artificial neural networks, Classification, Computer vision, Decision tree, Support vector machine |
| Status : Paper Accepted (Poster Presentation) |
